The perfect

• All East Caucasian languages have 1–4 forms resembling a perfect¹
• Many, but not all of these forms express evidentiality
• Chechen and Tsakhur employ a specialized auxiliary instead of a perfect to express evidentiality²

The forms in question look quite similar: most common are periphrastic constructions with a converb and a present tense copula.

##             
##              auxiliary copula no auxiliary person clitic
##   converb            3     26            8             8
##   participle         1      8            4             6
##   stem               0      2            0             0
##   unclear            0      0            4             0

The perfect can express tense-aspect and evidentiality at the same time, when the speaker infers from some result or consequence at speech time that a certain event occurred in the past.

If it is sufficiently grammaticalized as an evidential marker, the perfect can refer to an unwitnessed event in the past without the implication of some relevant consequence at the moment of speech. It is important to note that a single form can function as a full-fledged unwitnessed past in some contexts (most importantly narrative sequences), while retaining prototypical perfect functions in other contexts.

In some languages, the general past is considered a grammaticalized marker of direct evidentiality (witnessed past), in contrast with the perfect, which expresses indirect evidentiality (unwitnessed past). Since these direct forms also occur in the context of unwitnessed events, they could simply be neutral forms that carry witnessed overtones in particular contexts.

In the table below you can find all the forms I labeled as perfects or rather – “perfectoids”: forms that are very similar to the cross-linguistic perfect category, but which might not have current relevance (the prototypical meaning of the perfect) as their main function.³ (I am still working on linking the source references to their bibliographical information.)

Clitics

Specialized evidential morphemes in East Caucasian usually come in the form of clitic particles. Some of them attach to inflected verbs, while others are focus sensitive particles that can attach to any type of constituent (though they attach to the verb by default). These clitics often originate from verbs of speech, with some languages featuring petrified verb forms or even full verbs used as clitics. The vast majority of clitics mark reported speech. Inferential and general indirect particles are attested only in two languages.

Not all reported speech clitics are evidentials. Most of them are quotative particles, used to mark the boundaries of a quotation, rather than indicate the source of information per se. In East Caucasian a single morpheme can function both as a quotative and a reportative particle. The latter indicates that the information was based on hearsay. In my data, I labeled as “evidential” all of the clitics known to have a reportative function.

The data on clitics suffer from descriptive gaps. Particles are an underdescribed word class, and classification is inconsistent.

Small evidentiality survey

The small evidentiality survey consists of 11 questions, some of which elicit multiple sentences.⁴ The table below displays the number of sentences elicited from each speaker in conjunction with their language and idiom. This survey was exploratory in nature.⁵ Some of the Rutul results are discussed in Ferxees (2017).

## # A tibble: 6 x 4
## # Groups:   language, idiom [4]
##   language idiom  speaker sentences
##   <chr>    <chr>    <dbl>     <int>
## 1 Andi     Zilo         1         4
## 2 Andi     Zilo         2        16
## 3 Avar     ?            1        23
## 4 Dargwa   Itsari       1        11
## 5 Rutul    Kina         1        12
## 6 Rutul    Kina         2         8

The Perfect Questionnaire

The Perfect Questionnaire was completed with two speakers of Avar (Khunzakh dialect) and two speakers of Andi (Rikvani and Zilo dialect). The methodology considers one speaker sufficient, though more speakers can be consulted to “enhance reliability” (Dahl 2000, 800).

The speakers gave rather different answers, which could be due to dialectal variation (Andi) or sociolinguistic factors such as age or sex. One Avar speaker avoided using perfects (see the graph below), which I suspect might be due to his young age and lack of confidence in his own proficiency, though this is only speculation.

(The variable “not p” in the graph covers the general past tense (in this case: aorist) and its derivatives, such as the pluperfect with an aorist auxiliary, which is paradigmatically opposed to a pluperfect with a perfect auxiliary. The category “other” includes everything that is not perfect or aorist.)

Narrative test

The Perfect Questionnaire discussed above contains several questions meant to elicit a short narrative. These narratives are very similar, but each question proposes a slight change of perspective in terms of how long ago the events occurred, and whether the speaker supposedly experienced or witnessed them personally, or rather heard about them from someone else.

I found that these questions were very effective for eliciting the indirect evidential use of the perfect. So I created a test consisting of two very similar narratives (which are also very similar to the questions from the Perfect Questionnaire), but with a different evidential perspective. Narrative 1 retells an anecdote about the speaker’s grandmother (eliciting for indirect evidentiality). Narrative 2 relates a memory from personal experience (eliciting for direct evidentiality).

Read the test and the English translation of the questions here.

I conducted this test with speakers of different Andi dialects, mostly Rikvani and Zilo.

## # A tibble: 4 x 3
## # Groups:   language [1]
##   language idiom    speakers
##   <chr>    <chr>       <int>
## 1 Andi     Muni            1
## 2 Andi     Rikvani         5
## 3 Andi     Rushukha        1
## 4 Andi     Zilo            6

Most frequent across speakers were the aorist (which is the least marked past tense) and the perfect. As the graph below illustrates, these forms show the expected correlation with evidential perspective: aorist is more typical of the direct/witnessed context, while perfect is more frequent in the indirect/unwitnessed context.

Figure: Forms used in Bagvalal and Tsakhur texts by perspective

Now let us reduce the noise by looking only at the distribution of perfect with respect to general past in narrative sequences, since they form the backbone of a narrative.

Unfortunately, the Bagvalal and the Tsakhur sample are unequal in comparison, and neither is balanced (see the table below). Speakers are almost exclusively male, and some of them are overrepresented while others are underrepresented. This has to do with the fact that the material was collected during field trips, thus limiting the scope to a convenience sample of speakers willing to work with linguists.

## # A tibble: 4 x 6
## # Groups:   lang [2]
##   lang     sp_gender speakers sequences clauses av_length
##   <chr>    <chr>        <int>     <int>   <int>     <dbl>
## 1 Bagvalal f                2         6     150      25  
## 2 Bagvalal m                8        43    1221      28.4
## 3 Tsakhur  f                1         4      41      10.2
## 4 Tsakhur  m                7        13     876      67.4

Figure: Perfects and pasts by perspective in Bagvalal and Tsakhur

Both languages reveal an asymmetry between perfect and past that suggests a more neutral or less marked status of the latter. In addition, Bagvalal speakers seem to have a stronger preference for perfect in the indirect contexts. Note that this might reflect a sample bias, rather than a difference in the grammars of the languages.

Below are the absolute numbers that correspond to these graphs, as well as the results of a chi squared test for each of them, showing that the correlation between perspective and form is statistically significant.

Bagvalal: Statistical significance

##           
##             pf pst
##   direct    10 211
##   indirect 176  78

## 
##  Pearson's Chi-squared test with Yates' continuity correction
## 
## data:  bagv$persp and bagv$p
## X-squared = 205.36, df = 1, p-value < 2.2e-16

Tsakhur: Statistical significance

##           
##             pf pst
##   direct     3  97
##   indirect  51 114

## 
##  Pearson's Chi-squared test with Yates' continuity correction
## 
## data:  tsakh$persp and tsakh$p
## X-squared = 28.196, df = 1, p-value = 1.096e-07